Modify models

  • Release version: Australia
  • Updated March 12, 2026
  • 1 minute to read

    • Test and modify the Virtual Agent models so they more accurately predict user intents.

    Before you begin

    Role required: Chat Analytics Admin

    About this task

    Many ServiceNow products, such as ITSM, HRSD, and CSM include NLU models for Virtual Agent (VA) topics. The NLU plugin contains entities that are common across all products. You can modify those models. To learn how to train and test the models used for VA conversations, see Natural Language Understanding.

    Procedure

    1. Navigate to All > Conversational Interfaces > Analytics and select the NLU Prediction tab.
    2. Select View more .

      The Models page opens that displays the models that predict intents in your setup.

      NLU Workbench dashboard list of NLU models.

    3. Optional: If none of the models contain intents and utterances you'd like in your setup, select Create Model and follow the instructions in Create an NLU model.
    4. Click a model to display the intents the model predicts.

      NLU Workbench Model Intents tab.

    5. Optional: Click an intent to display the utterances associated with it.

      NLU Workbench Intent utterances tab.

    6. Optional: To create a new intent and add it to the model:
      1. Select New Intent.
        The Create a new intent dialog box opens.
      2. Enter the name of the new intent in Intent Name, a description of it in the Description field, and select Save.
        A page for your new intent opens.

        NLU Workbench intent utterances tab showing utterance added.

      3. Enter an utterance for your new intent and select Add.
        You can repeat this step multiple times to add multiple utterances.
    7. Optional: Import one or more intents by selecting Import Intents.
      The Import Intents dialog box opens, which displays the intents you can import.Import intents dialog box with SubmitRequest intent selected.
      1. Find the intent(s) to import by entering a search term, or by clicking an arrowhead to display the intents in a folder, and then selecting one or more check boxes.
      2. Select Import.
    8. Optional: If you added or imported intents:
      1. Select Train to add the intents to the model.
      2. Select Train to train the model.
        If successful, a banner at the top reads, The model has been successfully trained.
    9. To test all the utterances, select Test and enter a sentence a user might enter in a chat to see if NLU understands the intent of your sentence.

      NLU Workbench Intents tab and Test model tab displaying test utterance.

    10. If the models fail to predict the intent, add additional intents to the model, or utterances to the intents, as described in step 6.